Large Language Models and the Ethics of Generative AI

Hello students, welcome to Kwickprep. A chatbot answers your science doubt in perfect sentences. How does it do that, and should you believe every word? Today we will learn what a large language model is and how to prompt it well. Then we will build a small chatbot and learn to use AI responsibly.

Let us break the name into parts. LLM stands for Large Language Model. A language model is a program that predicts the next word in a piece of text. It is called large because it learns from a huge amount of text and stores billions of numbers, called parameters. Well known examples are GPT, Gemini, Claude and Llama.

An LLM does not read whole words the way we do. It first splits text into small pieces called tokens. A token can be a full word or only a part of a word. For example, unbelievable may be split into un, believ and able. The model then predicts the next token, again and again, to build its answer.

An LLM is trained in stages. In pre-training, it reads a huge amount of text and learns grammar, facts and patterns. In fine-tuning, it learns from example questions and good answers, so it starts following instructions. In the feedback stage, humans rate its replies, which makes it more helpful and safer.

Remember its limits before you trust it. An LLM does not truly understand the world like a person; it predicts likely words. It can hallucinate, which means it invents a fact and says it confidently. Its knowledge stops at the date its training data ended, unless it can search the web. And it can repeat unfair bias found in its training data.

A prompt is the instruction you give an LLM, and a good prompt gets a good answer. First, give a role, such as, act as a friendly physics teacher. Second, state the task exactly, such as, explain Newton's third law. Third, add context, such as, for a Class nine student in India. Fourth, ask for a format, such as, five bullet points under one hundred words.

Let us compare a weak prompt with a strong one. The weak prompt gives no role, while the strong one says, act as a teacher. The weak prompt says only, tell photosynthesis, while the strong one says, explain it simply. The strong prompt adds context, that the reader is in Class nine. It also fixes the format: five points and one example.

Pause and predict. Prompt A says only, write about Diwali. Prompt B says, write three lines on Diwali for a Class two greeting card. Which reply will be more useful? Prompt B, because it gives the length, the reader and the purpose. Also, if the first reply is not right, improve your prompt and ask again.

Now an advanced part, mainly for Class twelve students. An API, or application programming interface, is a set of rules that lets your program talk to another program. Our Python code will send messages to an LLM over the internet. The LLM service sends a reply back. To use it, you need an API key, which works like a password. Store it in an environment variable, and never type it in your code.

Here we use the OpenAI Python library, installed with pip install openai. The line bot equals OpenAI, with empty brackets, creates a client, and it reads your key from an environment variable named OpenAI API key. The variable model, in capitals, holds the model name, so check the provider's website for current names. The list chat stores every message, and the first one has the role system, which sets how the bot should behave.

Next, we write a function called ask. It sends the whole chat list and the model name to the service, and the reply comes back in the variable r. The text of the reply is inside r dot choices, index zero, dot message, dot content, and we store it in a. We add that reply to chat with the role assistant, so the bot remembers what it said. Finally, the function returns the reply.

Last, the loop that keeps the conversation going. While True repeats forever, and input reads the student's question into the variable q. If q double equals bye, break stops the loop. Otherwise we add the question to chat with the role user. Then we print Bot, followed by the reply from ask. Because chat keeps growing, the bot remembers earlier questions in the same conversation.

Now the ethics of Generative AI, starting with misinformation, which means false information that spreads. AI can write fake news that looks real and sounds convincing. It can also make deepfakes, which are fake photos, voices or videos of real people. So check important facts with trusted sources, like official websites or your textbook. And pause before you forward any shocking message on WhatsApp.

Next, copyright, which is the legal right a creator has over their original work. AI models learn from books, art and music made by real people, and many creators say this is unfair to them. The law on AI and copyright is still being decided in many countries, including India. So never pass off AI output as fully your own work. Say clearly when you used AI, and follow your school's rules for projects.

Finally, a few rules for responsible use. Never type passwords, Aadhaar numbers or other personal data into a chatbot. Use AI to understand a topic, not to copy homework answers. Never make fake images or messages that could hurt or insult someone. And remember, a human must always check the output and take responsibility for it.

Let us revise what we learned today. An LLM predicts the next token, after learning from a huge amount of text. A good prompt has a role, a task, context and a format. An API lets your Python code talk to an LLM, and the key must stay secret. And always check facts, respect copyright and use AI responsibly.